Portfolio Investment Based on Gene Expression Programming

A novel method of stock portfolio management by using technical indicators is proposed in this paper. The method hybridizes the consensus trading signals generated by the gene expression programming (GEP) proposed by Lee et al., and the portfolio redemption scheme proposed by Tsai et al. with our stock ranking functions. The indicators were used not only for trading, but also for selecting promising stocks into our portfolio. In order to search effective indicators for ranking stocks, the Pearson’s product moment correlation coefficient between the technical indicators with various parameters and one-day-ahead returns of the portfolio index (PI) are calculated. Then, seven significant indicators are found. Four weight functions W , k = 1, 2, 3, 4 are considered to aggregate these indicators. To get adaptive weights, the data of every three years are divided into the weighting interval (first year), aggregating interval (second year) and testing interval (third year). The experiment data set consists of 100 Taiwan stocks, from 2002/1/4 to 2015/12/31, containing a total of 3473 trading days. The highest average annualized return of our portfolio management method is 17.17% with the weight function combination (W ,WA). Furthermore, if the portfolio size and the redemption threshold are confined to 3 ≤ P ≤ 10 and 40% ≤ T ≤ 80%, respectively, the highest average annualized return is 17.26% with the weight function combination (W ,WA), which is better than the annualized returns of the buy-and-hold (BAH) rule (9.26%) and Lee’s method (11.05%). Keywords-gene expression programming; portfolio redemption scheme; stock investment; majority vote; technical indicator;